Introduction
Acute respiratory distress syndrome (ARDS) is a severe form of acute inflammatory lung injury, resulting in severe respiratory failure with marked hypoxia. ARDS is defined by the Berlin definition, which includes the clinical, radiographic, and laboratory aspects of ARDS:
- Acute onset of respiratory failure within 1 week of a known clinical insult
- Bilateral opacities on chest imaging, not fully explained by effusions, lung collapse, or nodules
- Respiratory failure not fully explained by cardiac failure or fluid overload
- Hypoxemia, as measured by the ratio of partial pressure arterial oxygen to fraction of inspired oxygen (PaO2/FiO2 ratio) ≤ 300 mm Hg
Despite decades of research, treatments for ARDS are still limited, consisting of lung-protective mechanical ventilation strategies and a few other supportive treatments. The mortality from ARDS remains high, between 35-46% depending on severity.
Prior to COVID, ARDS was reported to be present in 10% of all patients admitted to the ICU, and in 23% of patients requiring mechanical ventilation. However, because of the rapidly progressive nature of ARDS and the subjectivity involved in the interpretation of chest radiographs, the diagnosis of ARDS was frequently delayed or missed. One study reported that only about 60% of ARDS cases were recognized by clinicians, and only a fraction of patients received the appropriate treatments that were indicated for ARDS. As such, clinical decision support tools that can predict or identify ARDS would help clinicians recognize and treat ARDS in a timely manner.
Application Roadmap
RADS2 team developed “eARDS”, which is a machine learning algorithm for predicting ARDS development in an ICU population up to 12 hours before the patients satisfy the Berlin definition.